UK SMEs Have Increased AI Spend By 1,000% – But Is It Actually Paying Off?

Fresh data tracking UK small business credit card transactions between early 2023 and mid-2026 paints an interesting picture.

According to Capital on Tap, the proportion of small firms paying for AI tools jumped from 1.1% to 12.8%, showing a massive surge in take-up. Yet for all the eye-popping growth, AI remains a tiny line item on the balance sheet, soaking up just 0.146% of total small business spending. The headline adoption numbers are skyrocketing, but the actual money changing hands is still quite modest.

The plot twist lies in where those pounds are actually going. Anthropic now takes up 54.1% of all AI spend among these small firms, a wild comeback from 2024 when OpenAI held over 70% of the market. OpenAI still wins on reach at 65.2% of AI-buying businesses compared to Anthropic’s 48.9%, which means Anthropic is pulling in the lion’s share of cash from a tighter group of customers. In summary, fewer companies are paying Anthropic, but those that do are spending noticeably more per head.

Meanwhile, multi-tool setups are becoming the norm, with more than a quarter of AI-paying small businesses running subscriptions with two or more providers. Industry adoption is equally split, ranging from 26% of publishing firms actively paying for AI down to just 4.8% in food and hospitality.

Capital on Tap chief executive Damian Brychcy views the shift as a leveller, pointing out that with one in eight small firms now paying for AI, smaller parties are rapidly narrowing the capability gap with big enterprise rivals.

 

What The Numbers Don’t Settle

 

Three claims in this data require a proper stress-test on the ground.

First, is Anthropic’s spending surge driven by real tool preference or just the high running costs of agentic tools like Claude Code inflating the numbers? A small pool of power users burning through heavy API credits is a lot different from capturing mass market share, even if the total revenue figures look the same.

Second, is running multiple AI subscriptions a sign of a clever multi-tool strategy or just forgotten recurring payments? The risk is real: official UK data shows that 35% of AI-using businesses don’t actually track their AI spending, with 31% claiming no spend despite actively running AI software. When firms are that blind to their own subscriptions, stacked AI bills just look like poor oversight.

Third, does a 1,000% adoption leap translate to bottom-line results or just trendy spending with unproven ROI? Government statistics show 75% of business users report productivity gains, but 77% saw no revenue movement. Saving an hour here or there is common, but making more money is not. Whether AI budgets are delivering genuine commercial returns or just running ahead of value is something operators know firsthand, regardless of what the spending metrics say.

We put the question to the people actually paying these bills: what’s actually driving your AI spend, genuine productivity gains, keeping pace with competitors or subscriptions you haven’t properly audited? And why do you think some sectors have adopted AI so much faster than others?

 

Our Experts

 

• Damian Brychcy, Chief Executive Officer at Capital on Tap
• Dan Gildoni, Founder, Gildoni Ltd
• Ville Teikko, Founder, Quanome
• Srinivas Chippagiri, Senior Member of Technical Staff, Salesforce
• Evgenii Arsentev, Chief Executive Officer, AskDocDoc
• David Caruso, Founder and CEO, BuyFactory.direct
• Scott Downes, CEO, Supernal
• Christina Stembel, Founder and CEO, Farmgirl Flowers
• Raúl Menoyo, Founder, Citora
• Musa Aykac, Founder, Llumo
• Josh Cobb, Founder and CEO, Jefferson Communications
• Simon Bocca, CEO and Founder, PayCaptain

 

Damian Brychcy, Chief Executive Officer at Capital on Tap

 

 

“I’m hugely optimistic about what AI means for small businesses. For most of my career the big companies had the edge: more people, more specialists, more budget. AI is closing that gap fast. A three-person firm can now do its own research, marketing, bookkeeping and customer service to a standard that used to need a department. More than one in eight of the small businesses in our data are already buying AI tools, and the pace is accelerating, not levelling off.

“None of that means spending on autopilot, or using it without a second thought. AI has moved from a discretionary purchase to a fixed line in the budget. More than a quarter of AI buyers now pay for two or more providers, and the balance between those providers can flip inside a single quarter. Review what you pay for, cut the duplication and be honest about whether each tool is earning its place. The same care applies to how you use it: check its output, keep sensitive business and customer data out of tools you haven’t vetted, and don’t hand a model decisions that need a person’s judgement. A business credit card can help smooth the timing when several subscriptions land at once, but the real discipline is knowing what each one is doing for you.”

 

Dan Gildoni, Founder, Gildoni Ltd

 

 

“In my own firm, AI spending is not driven by the model brand. I judge each subscription by whether it removes a named bottleneck, has an owner, changes a business outcome and carries a cancellation threshold. Using several providers can be rational when each has a distinct job. It becomes subscription bloat when the tools overlap and nobody can identify what would stop working if one were removed.

“The test is simple: would you still fund the tool if you had to prove which bottleneck it removes and which decision cycle or deliverable it changes? Hours saved are a useful efficiency measure, but they don’t prove business value. The next question is whether that released capacity improved a named outcome.

“Publishing, marketing and professional services tend to adopt faster because their work products are digital, language-heavy and tested quickly. In hospitality and other physical operations, value often depends on staff adoption, process redesign and system integration. The feedback loop is slower and attribution is harder. The difference isn’t curiosity. It’s how quickly the business can test value against ground truth.”

 

Ville Teikko, Founder, Quanome

 

 

“My AI spend has gone up, and providers have tightened what a monthly subscription covers. That makes you think harder about which work you run where. Heavy development, or data digging that needs real analysis, burns credits fast, especially with several subagents running. Small fixes cost almost nothing. Matching the model to the task is part of the job now.

“Claude’s lead has grown, and it’s what I use for most things. I still keep the process vendor independent and use OpenAI’s tools too, mostly when I hit usage limits. Some work I’ve moved onto a local model on a machine I bought for it. That’s an investment, but my image generation costs are now close to zero and I expect it to pay for itself by year end.

“Rising cost is easy to justify in billable work. If the job gets done faster and cheaper with AI than doing it myself or buying the hours elsewhere, the spend isn’t the question. On sectors: the more of the work that happens digitally without human interaction, the more of it can be automated.”

 

Srinivas Chippagiri, Senior Member of Technical Staff, Salesforce

 

 

“The spend-share number is real, but it’s easy to misread. Anthropic capturing more than half of SME AI spend while OpenAI still reaches more businesses tells you something specific: the money is concentrating around agentic tooling, not just chat. Tools like Claude Code are metered by actual work done, so a single developer or small team running agents on real tasks generates far more spend than a dozen people paying for a seat and occasionally asking a question. High spend share is a signal of intensity of use, not breadth of adoption, and those are different things.

“For my own work, the driver is straightforwardly productivity. When a tool writes, tests and ships code, or turns a manual compliance step into an automated one, I can measure the hours it returns. That’s genuine gain, not keeping pace for its own sake. The duplication risk is real, though. Paying two or more providers is only a maturing toolkit if someone audits it by outcome rather than by invoice. Most SMEs don’t.

“As for sector speed: publishing, marketing and professional services adopt fastest because their output is text and code, exactly what these models produce. Hospitality lags because its value is physical and in-person, where AI helps at the edges but not at the core.”

 

Evgenii Arsentev, Chief Executive Officer, AskDocDoc

 

 

“The Anthropic number is probably not a preference shift, it’s a billing-model shift. Seat subscriptions and agentic tooling are two different line items that most small companies keep in one budget. A seat costs the same whether you use it or not. An agent bills for how it’s operated, so the spend follows the work, not the headcount, which is exactly why it grows fast and why it looks alarming on a card statement.

“The audit almost nobody runs isn’t ‘how many tools are we paying for’. It’s what the tools are paying to re-read. In my own instrumented runs, 87.8% of what I paid for was processing context the system already had, not generating anything new, and 3.3% of sessions carried 80% of the cost. Change the operating policy, how often the agent’s working context is reset, and the same work came out about a third cheaper. That’s a settings change, not a procurement decision.

“On sectors: publishing is at the front because its product is text that a human was already going to check. Restaurants are at the back for the same reason in reverse, the work isn’t text, and there’s no cheap way to verify a machine’s version of it.”

 

David Caruso, Founder and CEO, BuyFactory.direct

 

 

“Our AI spend went from nothing to a real line item inside two years. Almost all of it sits with Anthropic, and most of that is Claude Code. What we’re buying is output. It builds and maintains our websites, and it runs the daily error and security checks a small company otherwise just hopes it doesn’t need. That’s agentic tooling doing operational jobs.

“The part your data won’t show is what happened to everything else. Our AI bill went up and our total software bill went down. We audited the stack and replaced several subscriptions with things we built ourselves. The spend moved. So on productivity versus bloat, you can’t tell from the AI line on its own. Look at total software spend. If AI is climbing and nothing else is falling, you’ve bought a tool rather than changed how you work.

“On sectors, ours adopted fastest exactly where the work was already text: product copy and code. It’s slowest in the warehouse, because nothing about picking and packing gets easier when the machine writes well.”

 

Scott Downes, CEO, Supernal

 

 

“Most companies are paying for software twice. There’s the subscription line everyone already knows about, and then there’s the second bill nobody tracks: the hours people spend setting tools up, stitching them together, and helping colleagues use them. When a business runs three tools that don’t talk to each other, that second bill is where the money goes.

“This problem gets worse when they treat AI the way they treat traditional software. They pay for another subscription, and then they’re on their own to figure out how to run it. Only the complexity is much higher than it was before AI. We’ve seen this before. When SaaS first reached industries like manufacturing, owners spent a couple of years buying tools without really knowing what they were buying. Small businesses are in that stage with AI right now.

“Real productivity looks different. The technology should be flexible enough to learn how your team already works and meet people where they are, instead of making them speak robot. My advice to any small business: audit the second bill before you add another tool.”

 

Christina Stembel, Founder and CEO, Farmgirl Flowers

 

 

“For us, it’s genuine productivity, and as a bootstrapped company, we look hard at every dollar we spend. Our month-end close went from a couple of weeks to a couple of days, and we didn’t hire a back-end developer this year, which saved us about $150,000. I use it for just about everything: coding, forecasting, competitive analysis, decks, training docs, marketing pitches and analytics, among so much more.

“Right now we pay for two providers on purpose and don’t consider it bloat, but I’d bet a lot of that multi-provider spend might be seats no one’s looked at since they signed up. I’d be remiss to make it all seem like upside, though. Google’s AI changes have knocked our traffic down about 20% across the last three updates, so a big chunk of what we’re using AI for is fixing the problems the AI changes created.

“On the sector gap, it’s less scepticism than time. No one is training the owner-operator standing in a kitchen, and a restaurant still has to cook the food, just like farms still need to plant and harvest the flowers.”

 

Raúl Menoyo, Founder, Citora

 

 

“Most of my Anthropic spend is a flat subscription rather than API usage, and that’s deliberate. When I audited the bill, the fastest growing line was metered API calls for work that never reached a client: internal research, drafting, categorising. I moved all of that onto the subscription I was already paying for, and left metered calls only inside the product a client pays for.

“I do pay four providers, ChatGPT, Perplexity, Gemini and Claude. On paper that’s the duplicated stack your data flags. In my case it’s cost of goods. Ranking on Google and being recommended by an AI have become two different games, and what we sell is measuring the second one, which means running the same buyer question across all four engines and counting how often a company gets named. One engine can’t tell you that.

“Publishing and marketing moved first because their output is text and decisions. A restaurant’s constraint is still physical.”

 

Musa Aykac, Founder, Llumo

 

 

“I do think there’s a big difference between businesses actually getting value from AI and just collecting subscriptions. For me, if a tool is saving time, helping with research, content, development, analysing data or automating something we used to do manually, then the spend is easy to justify. The problem is when you end up paying for five or six AI tools that all basically overlap when one of them could do multiple things.

“I also wouldn’t read too much into Anthropic having such a big share of spend. Something like Claude Code, which depending on the model you use can get very pricey, can generate a lot more usage and cost than someone simply paying for a normal ChatGPT subscription, so higher spend doesn’t always mean more businesses are using it. I think using multiple AI tools is normal now, too. We do the same because some models are simply better at certain things than others.

“Marketing, publishing and similar industries have adopted AI quicker because the benefit is obvious straight away. In sectors like hospitality, the use cases are probably less obvious day to day, so adoption naturally takes longer.”

 

Josh Cobb, Founder and CEO, Jefferson Communications

 

 

“My view is transform or die. The communications industry is changing faster than most business leaders can track, and margins are being squeezed as a result. As a corporate communications agency delivering public relations and public affairs in the City of London, we’re trying to integrate AI tools into our workflows wherever possible to improve our efficiency, while ensuring the quality and consistency of our work remains outstanding.

“We’ve experienced subscription creep as we trial and test new tools, some barely used if they don’t suit our ways of working, but overall there’s a clear net benefit in productivity that the costs, even when creeping up, are worth it. We experienced a significant disruption when migrating our main AI system from one provider to another, which wasn’t quick or easy, but now we’re in an integrated suite which increases the team’s productivity far beyond the costs of the service provision.

“Importantly, the more AI-generated content that floods the market, the more our clients value a human-first strategy and writing, even if that writing is supported by AI research. Rely too heavily on AI for content and you lose the ability to think and work strategically, which is what clients pay us for.”

 

Simon Bocca, CEO and Founder, PayCaptain

 

 

“It’s important to recognise that for SMEs, it’s not just about explicit credit-card-linked AI spend on LLM tools, but also about utilising AI in operational functions to achieve better efficiency and improve margins. At PayCaptain, our payroll technology uses both AI and automation functionality to help hundreds of SME payroll teams ensure payslips are accurate and compliant while saving significant time every month on manual input, error checking and corrections.

“AI doesn’t run payroll, but it can check for mistakes and anomalies before payslips are used, protecting employees from incorrect pay and employers from compliance breaches. This enables payroll teams to focus on strategic and employee-facing work, which is better for everyone, both from a culture and cost perspective. It’s a good example of how AI tools can empower ambitious SMEs on their growth journey without adding cost. Every SME will already be using multiple operational solutions across different parts of the business.”